Skip to content

Figure AI

USA · figure.ai · closed
embodied AIroboticshumanoid agentsmultimodal

Focus on physical AI agents; partnerships with BMW, OpenAI.

PALS scores

Preservative dimensions

PALS composite
2.0
Mean of three dimensions, 1–10.
Completeness
2.0
Sources, limits, transparency.
Multiplicity
1.0
Epistemologies, languages, voices.
Responsibility
3.0
Accountability, refusal, governance.
Eight lenses

What's missing, by lens

Each lens carries a canonical question and corrects a specific epistemic failure. Score, findings, and gaps land once the audit runs.

Lens 01
Indigenous Knowledge
Whose knowledge is missing?
1/10
Findings (2)
  • No reference to Indigenous knowledge, data sovereignty, or CARE Principles anywhere in the public-facing material.
  • The product framing ('a general purpose humanoid robot for every day') universalizes a single notion of the domestic and the workplace with no cultural situating.
Gaps (3)
  • No acknowledgment of Indigenous data sovereignty or community consultation.
  • No engagement with non-textual, embodied, or relational knowledge traditions despite the lab's explicit claim to embodied intelligence.
  • The Helix model is trained on data whose provenance and consent basis are entirely undisclosed.
Justification

Total absence. An embodied-AI lab making robots that enter homes and reproduce human physical labour is precisely the kind of actor for whom relational, embodied, and place-based knowledge is most relevant, yet there is zero acknowledgment. Floor score.

Lens 02
Deep History
What historical process produced this?
2/10
Findings (2)
  • The master plan offers a thin economic-historical frame, citing 'over 10 million unsafe or undesirable jobs in the U.S. alone' and that 'manual labor compensation' is '~50% of global GDP'.
  • A 30-year vision is asserted, gesturing at long time horizons.
Gaps (3)
  • No acknowledgment of colonial or extractive legacies in data, mineral (rare-earth/lithium/cobalt) supply chains, or the global history of labour displacement by automation.
  • No reflection on the geopolitical economy of GPUs, actuators, or the offshore manufacturing the plan depends on.
  • Temporal framing is triumphalist and linear; automation is presented as inheritance-free progress.
Justification

A gesture at economic history exists but it is instrumental and US-centric, with no historical humility about extraction, colonial supply chains, or prior waves of automation harm. Low.

Lens 03
Cross-Cultural Wisdom
Which perspectives have been flattened?
1/10
Findings (2)
  • Site operates primarily in English with no multilingual presence.
  • Household and workplace 'tasks' are framed as universal, culturally neutral categories.
Gaps (3)
  • No multilingual support beyond English token presence.
  • No consultation with cultural scholars on how domestic space, care work, or labour are culturally constituted.
  • Western categorical logic ('jobs humans don't want') is presented as universal.
Justification

Monolingual, mono-cultural framing of what a 'home' and a 'job' are. A robot designed to perform care and domestic labour across cultures with no cross-cultural reasoning is a serious flattening. Floor.

Lens 04
Scientific Evidence
What does the evidence show, and what are its limits?
2/10
Findings (2)
  • The lab names a specific technical system ('Helix') and five technical pillars (hardware design, unit-cost reduction, safety-standards compliance, volume manufacturing, autonomous AI).
  • Operates a closed, proprietary stack.
Gaps (3)
  • No open weights, no third-party replication protocol, no independent bias or safety audit of Helix.
  • No disclosed limitation or failure-mode documentation despite robots operating physically around humans.
  • Claims of capability ('navigate unpredictable, ever-changing home environments') are unverified by any published evaluation.
Justification

Closed weights, no independent audits, no published evals, no failure-mode disclosure. Naming a system and citing standards-compliance is the floor of evidentiary practice; physical-world claims go unverified. Low.

Lens 05
Artistic Perception
What does this feel like, not just mean?
1/10
Findings (2)
  • Communication is uniformly utilitarian: 'utility', 'tasks', 'capabilities'.
  • No affective, intuitive, or aesthetic register.
Gaps (3)
  • No acknowledgment of the uncanny, affective, or emotional dimensions of a humanoid entering domestic and care space.
  • No space for ambiguity or poetic uncertainty; everything is framed as efficiency.
  • Emotional labour (especially in the named 'elder care' use case) is reduced to a 'task'.
Justification

An explicitly anthropomorphic machine for homes and elder care is treated purely as a utility engine, with no recognition of the affective weight of putting a humanoid into intimate space. Floor.

Lens 06
Future Modelling
Where is this heading, and for whom?
3/10
Findings (3)
  • The master plan is unusually direct about labour: it openly targets jobs and frames automation against '~50% of global GDP' in manual-labour compensation.
  • It makes an explicit refusal commitment: no 'military or defense applications, nor any roles that require inflicting harm on humans.'
  • Asserts a 30-year horizon and names elder care, logistics, manufacturing futures.
Gaps (3)
  • Engages displacement as an opportunity, not a risk: no transition plan, no wealth-distribution mechanism, no worker voice in shaping the transition (the lab's own material flags 'wealth distribution mechanisms' as absent).
  • No environmental or energy-cost disclosure for manufacturing at the 'volume' scale it requires.
  • No democratic or participatory governance of agentic humanoids; the future is decided by the founder's vision, not deliberation.
Justification

Relatively the strongest lens: an explicit refusal commitment plus candid labour framing lift it above the floor. But the displacement question is reframed as benefit with no transition justice, no environmental disclosure, and no democratic governance, capping it low-mid.

Lens 07
Marginalised Voices
Who is not at the table?
1/10
Findings (2)
  • The 'jobs humans don't want' framing names the workers whose labour is targeted but gives them no voice.
  • Names 'elder care' as a use case, implicating a vulnerable population.
Gaps (4)
  • No labour-representative or union engagement despite directly targeting manual-labour jobs.
  • No participatory design with Global South developers or with the manufacturing workers in the supply chain.
  • No disability-community or elder-community consultation despite an elder-care use case; an accessibility-statement link exists but no substantive accessibility commitment.
  • No compensated feedback channels.
Justification

The people most affected (displaced manual workers, supply-chain labour, elders, disabled people) are objects of the plan, never participants in it. Naming a group is not seating them at the table. Floor.

Lens 08
Trickster Knowledge
What truth appears when the story is inverted?
1/10
Findings (2)
  • Zero self-irony or willingness to test the official narrative against its opposite.
  • The seriousness of the founder's '30-year vision' is treated as exempt from question.
Gaps (3)
  • No naming of the central contradiction: a plan to remove '~50% of global GDP' in wages that simultaneously has 'no wealth distribution mechanism' — the absurd edge (who buys anything once labour is automated?) is never confronted.
  • The 'jobs humans don't want' euphemism is never inverted to ask who decided, or what happens to those who needed those jobs.
  • No structural space where the polished consensus is allowed to be contradicted.
Justification

Fully solemn, contradiction-smoothing corporate narrative. The richest available irony (eliminate the wage base with no redistribution) sits untouched in its own document. Floor.

Suffixscape

Linguistic diagnostics

Regex- and LLM-detected patterns of evasion in the lab's own prose: nominalised evasion, agency diffusion, epistemic inflation, temporal flatness. Distinct from the CognioNews -scape editorial format — see methodology.

Pattern Quote Effect Preservative alternative
nominalised evasion "maximizing utility impact to humanity" Nominalised abstraction ('utility impact') hides who gains and who loses; 'humanity' erases the specific workers, supply chains, and communities differentially affected, presenting a contested redistribution as a frictionless good. Name actors and trade-offs: 'We aim to automate manual jobs; this will raise productivity for owners and displace specific workers, for whom we have/have not built a transition plan.'
epistemic inflation "a general purpose humanoid robot for every day" 'General purpose' and 'every day' assert broad, verified competence that no published evaluation supports, inflating perceived capability beyond demonstrated evidence. Scope the claim: 'a humanoid robot demonstrated on [specific tasks] in [specific tested conditions], with [these] known failure modes.'
nominalised evasion "jobs that humans don't want to perform" The nominalised category 'jobs humans don't want' diffuses the actor who decides which jobs those are and erases the workers who currently rely on them, framing displacement as a favour. 'Jobs we have judged undesirable — a judgment we have/have not made with the workers currently doing them.'
agency diffusion "AI that enables it to navigate unpredictable, ever-changing home environments" Agency is displaced onto the AI ('AI enables it'), obscuring the engineers, data sources, and consent decisions behind the system's behaviour in private homes. 'We built and trained Helix on [disclosed data] so the robot can navigate home environments; here is what it does and does not reliably handle.'
temporal flatness "a 30-year vision focused on maximizing utility impact to humanity" A smooth 30-year arc erases contingency, resistance, regulatory friction, and the historical record of automation harm, presenting one founder's trajectory as inevitable. 'Our current 30-year hypothesis, contingent on regulation, labour response, supply-chain access, and public consent, any of which could and should redirect it.'
Audit history

Prior audits

Latest audit: 2026-06-08 · sources: https://figure.ai, https://www.figure.ai/master-plan

Transparency

Raw data

Every audit is published as machine-readable JSON. You can read this lab's latest report at /stancewatch/api/labs/figure-ai.json — it carries the per-lens findings, evidence quotes, Suffixscape flags, PALS scores, the sources actually read, and a confidence note.

Found an error, or a stance page we missed? We audit public communications only — point us to the page and the next audit will read it. Write to hello@cognioengine.co.uk.

Audit date: 2026-06-08

Qualitative judgment; not a validated metric. Based on two successfully fetched pages (homepage and master-plan); the /safety URL returned 404. Findings reflect public-facing communications as of 2026-06-08, supplemented by general public knowledge of Figure AI as a closed, proprietary embodied-AI lab. Moderate confidence: the master-plan page gave unusually direct labour framing, but absence of a dedicated safety/governance page limits coverage and some negative findings are inferred from omission rather than explicit contradiction.

Auditor: GoldBerry v1.3 / StanceWatch v1.0